Machine Learning-Based Fatigue Detection From EEG Signals: A Methodological Pilot Framework With Potential For Nursing Applications

Abstract

Objective: Fatigue is a significant occupational health issue that adversely affects attention and decision-making. This study aimed to objectively predict fatigue levels using EEG signals and machine learning models and to develop a physiologically based assessment framework with potential future applications in nursing. Materials and Methods: Time-domain, frequency-domain, entropy, Hjorth, and wavelet features were extracted from EEG data collected from 12 participants aged 21–40 years, with fatigue levels labeled using various scales. Results: Classification performance varied across validation strategies. Under 5-fold cross-validation, Random Forest achieved the highest performance (accuracy=1.00, macro-F1=1.00), while KNN and NB also demonstrated strong performance. However, under LOOCV, sensitivity, precision, and macro-F1 values decreased despite relatively high accuracy. LOSO validation yielded more balanced results, with Random Forest maintaining the best performance (accuracy=0.97, macro-F1=0.91). Conclusion: EEG-derived features provide valuable information for fatigue assessment. Although high performance was achieved under certain validation strategies, subject-independent evaluation revealed more realistic performance levels. The proposed framework offers a practical approach to fatigue monitoring. Future studies should focus on larger, domain-specific datasets and multimodal integration to enhance generalizability.